Search papers, labs, and topics across Lattice.
Institute of Automation, Chinese Academy of Sciences
1
0
3
8
LLMs can be made to reason much better by directly optimizing their pre-training output distribution, even before fine-tuning on specific tasks.
